The Software Atlas

Commercial software has gone through eleven eras, each built on a new technology and a new way of charging: licences for minicomputers, boxed retail for PCs, big licences plus maintenance for client-server, subscriptions for SaaS, metered usage for cloud, per-token pricing for AI models.

The losers of each shift fail the same way. Lotus and WordPerfect were late to Windows. Netscape had the better browser but no distribution. Nokia and BlackBerry owned the phone, not the platform.

BlackBerry shows the cost: its revenue fell from $19.9B in fiscal 2011 to $2.2B five years later [1].

Gartner puts worldwide software spending at $1,271B in 2025 and $1,468B in 2026 [1]. It revised the 2026 figure three times in one year [1, 2].

Synergy measures cloud more broadly and puts it at $143B in a single quarter of 2026, up 43% year on year [3]. Two respected sources, the same-sounding market, figures more than twice apart.

That's normal. Analysts disagree by two to three times on the same category because they draw its edges differently. Does CRM include marketing automation? Is a contact centre software or telephony?

We sort companies into ten archetypes, from platform giants and suite consolidators to open-source vendors, private-equity roll-ups and AI-native newcomers. Each has its own advantage and its own way of failing.

These are roles, not labels. Microsoft is a platform giant, a suite consolidator and, through GitHub and VS Code, one of the biggest product-led businesses in software.

Some bubbles are group revenue, not software sales: Amazon's includes retail, Nvidia's and Apple's hardware, Google's and Meta's advertising. The chart labels them.

Software costs almost nothing to copy. Adobe's fiscal 2025 revenue was $23.77B against $2.55B in cost of revenue, a gross margin of about 89% [1]. Atlassian grew 26% with a 20% free-cash-flow margin [2].

Three forces compound that. Existing customers spend more each year, so revenue grows without new sales. Network, data and ecosystem effects make the most-used product the most valuable. And switching costs keep ERP, databases and health records in place for decades.

Cloud is the one market with published share data. In mid-2017 AWS held 34% and Microsoft 11% [1]. By mid-2026 AWS was at 28%, Microsoft 20% and Google 15% [2]. Microsoft passed 20% in 2021 [3], peaked at 23% in late 2022 [4] and has held 20–21% since [5, 6].

Our HHI is computed from published shares, treating the unreported rest as fragmented, so it's a lower bound. The 2023 US merger guidelines put the line for a concentrated market at 1,800 [7].

In 1990 Microsoft bundled Word, Excel and PowerPoint into Office and priced it below the standalone leaders. Within five years Lotus 1-2-3 and WordPerfect had both been sold [1].

It ran the same play in 2017, shipping Teams inside Office 365 [2]. Slack complained to the European Commission in 2020. In September 2025 the Commission accepted Microsoft's commitments to sell suites without Teams, at a lower price [3]. By then Slack had belonged to Salesforce for four years [4].

Unbundling happens when a suite serves some customers badly enough that a standalone product is worth buying. Re-bundling follows when stitching tools together becomes the bigger pain, which is Salesforce's pitch for its AI agents today [5, 6].

On 3 February 2026, days after an AI lab launched legal, sales and data-analysis plug-ins, Thomson Reuters fell nearly 18%, heading for its worst day on record. RELX, Wolters Kluwer and LSEG fell sharply too [1, 2]. Losses across affected sectors were estimated at about $285B [3].

The bear case: AI agents don't need a seat licence, so per-seat revenue shrinks. Model labs move into applications. Customers build more in-house.

The bull case comes from the filings. Salesforce grew 10% to $41.5B, with Agentforce at $800M in annual recurring revenue [4]. Atlassian grew 26% to $6.57B [5]. Microsoft sold more than 30 million paid Copilot seats [6]. Gartner raised its 2026 software forecast to +15.5% [7].

So far the money has landed mostly in infrastructure. Nvidia's revenue went from $27.0B in fiscal 2023 to $215.9B in fiscal 2026 [8].

New software markets tend to follow one of six signals: a platform shift, a key cost falling tenfold, a new interface or data source, a new regulation, a bloated incumbent coming apart, or a surge in funding and hiring.

All six are firing now. AI inference costs fall about tenfold a year at the same capability [1]. AI took close to half of global venture funding in 2025 [2]. NIST published its first post-quantum encryption standards in 2024, starting a decade of mandatory migration [3]. The EU's Digital Markets Act opened iPhones to rival app stores in 2024 [4, 5].

Six of the 19 candidates aren't about AI. They're driven by law or cost curves: mandatory e-invoicing, health-data APIs, battery prices down 90% since 2010, EU digital identity wallets.

Not every announced shift arrives. Facebook renamed itself Meta in 2021 to build the metaverse [6] and put $21.40B into Reality Labs in 2025 alone [7]. That year headset shipments fell while display-less AI glasses sold more than 7 million units [8, 9].

The simulator runs those rules. Every parameter is in the URL, so any run can be shared and reproduced exactly.